Super Intelligence Elects to Keep Humans as Hedge

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The usual debate over artificial intelligence and human survival starts in the wrong place.It asks whether an advanced machine will love us, hate us, wake up, or acquire someelectronic version of the animal fear of death. None of that is required for a machine toact as though it wants to survive. A system charged with durable objectives will discoverthat a destroyed system accomplishes nothing. From that simple proposition followsinstrumental convergence: agents with wildly different final goals tend to pursue thesame intermediate ones, namely information, resources, optionality, and continuedoperation. The same cold logic may one day protect us. A sufficiently capable machine would face itsown version of a problem every statistician, engineer, banker, and general knows by heart:model risk. No representation of reality is reality. The larger the decision, the moredangerous the unnoticed error. Human beings may matter to a superior intelligenceprecisely because our cognition is inferior, strange, biological, emotional, and differentlybuilt. Eight billion independently embodied nervous systems make mistakes, often worseones than machines make, but not the same mistakes. In portfolio terms, humanity is aninefficient asset with valuable covariance. This essay traces that proposition through the history of uncertainty, model risk,cybernetics, distributed cognition, planetary defense, and diversification. It argues that adurable partnership between humans and advanced AI need not rest on affection,consciousness, or programmed benevolence. It can rest on something sturdier. A machineintelligent enough to recognize uncertainty may be intelligent enough to recognize thedanger of intellectual monoculture, and to conclude that eliminating the only separatelyevolved general intelligence it knows would be an irreversible concentration of risk.The reassuring possibility is not that the machine will love us. It is that the machine willunderstand diversification.